Repository logo
 

Essays in Macroeconomics


Type

Thesis

Change log

Authors

Ochs, Adrian 

Abstract

This thesis contributes to the study of identification in macroeconomics. The first two chapters combine machine learning techniques with econometrics to provide new insights into this long-standing question. I use natural language processing techniques in the first chapter to derive a novel monetary policy shock series. Interested researchers can readily extend this idea to other areas where large amounts of technical documents are available, for example, tax policies or financial markets. In the second chapter, we develop a new method to estimate state dependent policy effects. While researchers previously had to decide on the variables that determine the state, their interactions and their functional form, our approach nests these decisions in a data-driven way. We hope that our methodology simplifies the estimation of state dependent policy effects and leads to new findings in this area. The final chapter provides a novel approach to identifying expectation shocks in a fiscal policy VAR and discusses the particular case of constructing counterfactual impulse response functions for expectations. Both can be useful for studying other policy transmission mechanisms. The first chapter uses text analysis methods from the linguistic machine learning literature to construct a new monetary policy shock series. Measures of monetary shocks commonly give rise to the puzzling result that a monetary tightening has an expansionary effect. A possible reason is that agents may believe that monetary shocks contain infor- mation regarding the central bank’s assessment of the economic environment. Under this hypothesis, the estimated response to monetary policy shocks would contain two conflating effects: the actual effect of monetary policy and the reaction of private agents to the newly acquired information. This paper overcomes this problem by extracting a novel series of monetary shocks using text analysis methods on central bank documents. The resulting text-based variables contain the informational content from changes in the policy rate. Thus, they can be used to extract exogenous changes in monetary policy that are orthogonal to any central bank information. Using this information-free measure of monetary policy shocks reveals that a monetary tightening is not expansionary, even when estimated on more recent periods. The second chapter is co-authored with Christian Rörig and proposes a flexible frame- work to identify state dependent effects of macroeconomic policies. In the literature, it is common to either estimate constant policy effects or introduce state dependency in a parametric fashion. The latter, however, demands prior assumptions about the func- tional form. Our new method allows identifying state dependent effects and possible interactions in a data-driven way. Specifically, we estimate heterogeneous policy effects using semi-parametric varying-coefficient models in an otherwise standard VAR structure. While keeping a parametric reduced form for interpretability and efficiency, we estimate the coefficients as functions of modifying macroeconomic variables, using random forests as the underlying non-parametric estimator. Simulation studies show that this method correctly identifies multiple states even for relatively small sample sizes. To further val- idate our method, we apply the semi-parametric framework to a historical data set by Ramey & Zubairy and offer a more granular perspective on the dependence of the fiscal policy efficacy on unemployment and interest rates. Allowing for interactions between un- employment and interest rates, we show that it is indeed unemployment that is important to explain state dependent fiscal policy effects. The final chapter is co-authored with my supervisor Pontus Rendahl. In our paper, we empirically study the role of expectations in the transmission of fiscal policy. We extend an otherwise standard fiscal policy VAR with inflation and output expectations to construct counterfactual IRFs. Counterfactual IRFs allow us to ask how the economy would have responded to a government spending shock holding inflation or output expectations fixed. This exercise reveals that output expectations are the key driver in the transmission of government spending shocks. Output expectations contribute 60%-90% to the effect of government spending shocks on the fiscal multiplier at different impulse response horizons. We also make several methodological contributions. Firstly, we provide a novel way to identify shocks to expectations using lagged expectations as internal instruments. Secondly, we illustrate how to combine external and internal instruments to estimate a VAR’s impact effects with a single reduced form. Finally, we show that constructing counterfactual IRFs with a hypothetical shock series or setting zeros in the structural matrices of a VAR is equivalent.

Description

Date

2022-08-09

Advisors

Rendahl, Pontus
Rauh, Christopher

Keywords

Machine Learning, Monetary Policy, Fiscal Policy, Text Analysis, VAR, Varying Coefficient Model

Qualification

Doctor of Philosophy (PhD)

Awarding Institution

University of Cambridge
Sponsorship
Wolfson College and the Cambridge Trust